Data Scaling Laws in Imitation Learning for Robotic Manipulation

Fanqi Lin*, Yingdong Hu*, Pingyue Sheng, Chuan Wen, Jiacheng You, Yang Gao

Project Page / Paper / Code / Summary / Models / Processed Dataset / Raw GoPro Videos

• Fundamental scaling laws: policy generalization performance to novel environments and objects follows a power-law relationship with the number of training environment-object pairs.
• Efficient data collection strategy: we gathered enough data to train policies with ~90% success on two tasks in any novel scenarios in one afternoon with 4 data collectors.

CoPa: General Robotic Manipulation through Spatial Constraints of Parts with Foundation Models

Haoxu Huang*, Fanqi Lin*, Yingdong Hu, Shengjie Wang, Yang Gao

Project Page / Paper / Code / Summary

• CoPa is a novel framework that incorporates common sense knowledge embedded within foundation VLMs into low-level robotic manipulation tasks.
• CoPa is capable of handling diverse open-set instructions and objects in a zero-training manner.

Look Before You Leap: Unveiling the Power of GPT-4V in Robotic Vision-Language Planning

Yingdong Hu*, Fanqi Lin*, Tong Zhang, Li Yi, Yang Gao

Project Page / Paper / Video / Summary

• We introduce ViLa, a novel approach for long-horizon robotic planning that leverages GPT-4V to generate a sequence of actionable steps.
• ViLa empowers robots to execute complex tasks with a profound understanding of the visual world.

TiZero: Mastering multi-agent football with curriculum learning and self-play

Fanqi Lin, Shiyu Huang, Tim Pearce, Wenze Chen, Wei-Wei Tu

Paper / Code / Match Video

We introduce TiZero, a self-evolving multi-agent system that combines the JRPO algorithm with curriculum self-play in a large-scale distributed training framework, achieving state-of-the-art performance in the GFootball 11 vs. 11 game.